GPT OSS Safeguard 20B

GPT OSS Safeguard 20B is part of OpenAI's gpt-oss-safeguard family, introduced as a research preview of open-weight reasoning models built specifically for safety classification work. Rather than baking a fixed policy into the weights, the model interprets a developer-provided safety policy at inference time, using chain-of-thought reasoning to classify user messages, completions, and full conversations against whatever rules the operator supplies, which makes it easier to iterate on policies without retraining.

As the smaller of the two release sizes, GPT OSS Safeguard 20B is a fine-tuned derivative of OpenAI's earlier gpt-oss open models and ships under the same permissive Apache 2.0 license, with weights distributed through Hugging Face. The transparent chain-of-thought lets developers audit how a classification decision was reached, offering a practical middle ground for teams that need customizable, reviewable content moderation without building and labeling their own classifier from scratch.

OpenRouteropenai/gpt-oss-safeguard-20bgpt-oss

Open provider page Provider docs

Quick Info

Provider: OpenRouter
Model key: openai/gpt-oss-safeguard-20b
Release date: Oct 29, 2025
Last updated: Oct 29, 2025
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $0.075
Output token cost: $0.30

Limits

Output tokens: 65,536 tokens
Context window: 131,072 tokens

Transparent token rates

Compare GPT OSS Safeguard 20B pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about GPT OSS Safeguard 20B

Merge Gateway

CoverageAnalysis

CNBC's October 29, 2025 report covers the same gpt-oss-safeguard launch, confirming the explicit names gpt-oss-safeguard-120b and gpt-oss-safeguard-20b as fine-tuned versions of OpenAI's gpt-oss models released earlier in August 2025. The article notes these are open-weight models whose parameters are publicly availabl The reporting adds partnership and testing context not present in the primary announcement: OpenAI developed gpt-oss-safeguard in collaboration with Robust Open Online Safety Tools (ROOST), an organization focused on safety infrastructure for AI, and Discord and SafetyKit also helped test the models. Example use cases

OpenRouter

CoverageRelease Notes

LiteLLM v1.82.0 ships day-0 support for routing gpt-oss-safeguard-20b via Groq, listed as groq/openai/gpt-oss-safeguard-20b with a 131K context window at $0.075 per 1M input tokens and $0.30 per 1M output tokens, and explicitly flagged for guardrail inference. Inclusion in the v1.82.0 changelog matters for developers b The same release also broadens guardrail ecosystem support with realtime /v1/realtime WebSocket guardrails, new provider integrations (Noma v2, Lakera v2 post-call, Singapore PDPA + MAS regulatory policies, employment-discrimination blockers, code-execution blocker), and guardrail policy versioning, while routing /v1/m

OpenRouter

Coverage

An OpenRouter detail page (served from a mirror host) for OpenAI: gpt-oss-safeguard-20b exposes it via a unified API with a 131K context window, $0.075 per 1M input tokens and $0.30 per 1M output tokens, and a release date of October 29, 2025. The description matches OpenAI's positioning of the model as a 21B-parameter Under the routing configuration shown, Groq is the sole hosting provider, with a P50 latency of about 0.21 seconds and roughly 380 tokens per second throughput, 100% uptime over the measured window, and a ~55.94% cache hit rate that brings weighted-average effective prices to $0.05353/M input and $0.2976/M output. Tool

OpenRouter

Official sourceComparison

Compare gpt-oss-safeguard-20b from OpenAI to other AI models on key metrics including benchmarks, price, context length, and other model features.